Use new Google voices in TTS model and voices

[gianpaj]

gemini-2.5-flash-preview-tts
https://ai.google.dev/gemini-api/docs/speech-generation

https://aistudio.google.com/u/1/generate-speech

[RachidAFL]

did you manage to use the Multi-speaker audio via API ?

[aidivas]

Hey do you have thex2.5 api working?? If yes any code changes to podcastfy thx

[RachidAFL]

Hey do you have thex2.5 api working?? If yes any code changes to podcastfy thx

it think it works you just need to get the api from here:
https://aistudio.google.com/app/generate-speech

and then try to integrate it with the script. it works

[aidivas]

@gianpanj thanks! I will test your config…while you wait for the PR..

[aidivas]

hey all - not sure why theres a PR for this AFAIK you can just drop the LLM model name into the config.yaml file?

Test with Gemini flash 2.5 today and it works BUT ignores my custom chunking paramaters - so rolled back to gemini 1.5 pro for now

Using just these lines in config.yaml
content_generator:
llm_model: "gemini-1.5-pro-latest"
meta_llm_model: "gemini-1.5-pro-latest"
max_output_tokens: 8192

How the Code Works

Looking at content_generator.py, the model is loaded from:

1. Priority 1: model_name parameter passed to ContentGenerator.__init__()
2. Priority 2: config.yaml under content_generator.llm_model
3. Default: "gemini-1.5-pro-latest"

The code in content_generator.py line 740:

if not model_name:
    model_name = self.content_generator_config.get("llm_model")

Looking at `content_generator.py` line 78:

```python
elif "gemini" in self.model_name.lower():
    self.llm = ChatGoogleGenerativeAI(
        api_key=os.environ["GEMINI_API_KEY"],
        model=model_name,  # This passes your model name directly
        max_output_tokens=max_output_tokens,
        **common_params,
    )

@gianpaj - i did look at your PR but not sure why you added your changes into conversation_config.yaml instead of config.yaml

and did you replace the default model for gemini mulit TTS with the new flash preview???? Thanks
 geminimulti:
    default_voices:
      question: "R"
      answer: "S"
      model: "en-US-Studio-MultiSpeaker"
      question: "en-US-Kore-Neural"
      answer: "en-US-Zephyr-Neural"
      model: "gemini-2.5-flash-preview-tts"

Seems some confusion between LLM Models and TTS models in your PR or is the PR to replace this     model: "en-US-Studio-MultiSpeaker"?? thanks

[aidivas]

See here #271